主 题：Inference on ROC and Average Precision Based on Methods of Quantiles
主讲人：张朋
主持人：张术林博士
时 间：2013年12月10日上午10：30-11：30
地 点：通博楼B座212会议室
主办单位：统计学院 统计研究中心 科研处
主讲人简介：
张朋，男，2006年6月毕业于加拿大滑铁卢大学，获得博士学位。现任浙江大学数学系统计研究所特聘研究员，主要研究方向为：混合模型，关联数据分析，数据挖掘，测量误差模型
内容提要：
The objective of a rare target detection problem is to identify the rare targets as early as possible. Recall, precision and average precision (AP) are three popular performance measures for evaluating different detection methods. The AP is found also a good performance measure in medical screening test and diagnostic test, similar as the receiver operating characteristic (ROC) curve but focusing more on earlier cases. A popular assumption for the underlying distributions of the decision variables is the binormal distribution. However, the existing maximum likelihood (ML) methods for estimating the normal parameters require the actual observations. We propose a method of quantile and another ML method based on quantiles of normal distributions that rely on ranking only. Simulation studies show the good performance of these approaches.
